psylligent / sglang
Code health
This codebase scores 5.9 out of 10 for code health, which we rate fair. It also scores maintainability 6.3 and static performance 9.8 out of 10. The three are scored separately and never blended into one number. Risk is concentrated, as it usually is: 1,446 of 9,383 files are git hotspots, and they average 3.9 — which is where the fixes pay off most.
Do next
1 thing worth doing in the week to Sep 29, the last indexed commit, 1 of them now.
Now
- Critical
Clean up what this week's commits left: 27 serious findings in 14 files
11 commits in the last 7 days added or worsened them, and they are still open. Start with
experimental/sgl-router/src/state/kv_events/index.rs, which has 5 critical.- serious findings
- 27
- critical
- 8
- files
- 14
Large effort
Improve what Repowise can see
Add a test coverage report
Without one, Repowise cannot tell tested code from untested code, so every test-related action says “unknown”.
- line coverage
- Unknown, no report
Small effort
- Lines of code
- 2.2M
Recent activity
All commitsCommits
- 46beacf1[grpc rust] Define semantic frontend contract (#41709)1d ago
- 77a81996[PD] Preserve bootstrap metadata in native Messages requests (#39711)1d ago
- 3c4653b6[Metrics] Fix PD latency histogram accounting (#39706)1d ago
- f884231f[Docs] Add IQuestLab card logo to cookbook landing page (#41746)1d ago
- 51ace483[Metrics] Add request-level TPOT histogram (sglang:request_time_per_output_token_seconds) (#40275)1d ago
- 9a16e474chore: grant jain-ria CI permissions (#41728)1d ago
Decisions
- Reroute Elastic-EP scale writers to get_context().override via ServerArgs.overrideproposed · 1d ago
- Keep live-shadowed topology sizes reading from server_argsproposed · 1d ago
- Keep ParallelContext.getattr dynamo-traceable and gate helpers inside compiled forwardsproposed · 1d ago
- Runtime config readers use resolved bags, not operator input recordproposed · 1d ago
- Route config-leaf reads in parallel namespace through get_parallel()proposed · 1d ago
- Represent explicit None vs unset in environment handlingproposed · 1d ago
Everything we found
48,501 open
Every detector's open count by area, each with its worst example. An inventory, not a to-do list: the actions above are the part worth doing.
- worst item is critical severity. Code health33,852Untested hotspot · python/sglang/multimodal_gen/runtime/models/dits/ernie_image.py25 findings · 10.0 deducted
- worst item is medium severity. Refactoring6,854execute_mcp_tools+1.5 health, S effort · sgl-model-gateway/src/routers/grpc/harmony/responses/execution.rs
- worst item is medium severity. Doc drift143docs/cookbook/autoregressive/DeepSeek/DeepSeek-V4.mdxLink points at #3-4-dspark-speculative-decoding, which docs/cookbook/autoregressive/DeepSeek/DeepSeek-V4.mdx no longer declares.
- worst item is medium severity. Decisions1,458python/sglang/srt/server_args.py12 awaiting review · 1446 hotspots ungoverned
- worst item is medium severity. Ownership5,854python/sglang/srt/runtime_context.py95% single-owner concentration, 96 commits in 90d
- worst item is low severity. Dead code340examples/frontend_language/quick_start/anthropic_example_complete.pyUnreachable symbol (40 lines)
Composition
Open the graphExplore this codebase
- Docs12,868 pages across 15 modules
- ChatAsk this codebase a question and get an answer with its sources.
- Files9,383 files with per-file docs, health and history
- ArchitectureDependency graph, layers, and 48 entry points
- Code healthPer-file scores, 42,329 open findings, coverage and refactoring targets
- Knowledge graphEntities, communities, and the paths between them
- Change couplingFiles that keep changing together, mined from commit history
- CommitsChange-risk ranked history with agent provenance
- ContributorsBus factor, per-file maintainers, and the human/agent split
- StatsSize class, origin, lifetime churn, rhythm and records
- CostsWhat indexing this snapshot cost, by model and by run.